install: Azure Linux needs the libc headers named, or the build dies later - #70
Merged
Conversation
…later
Found by the distro matrix, and it is exactly the failure this suite exists for: pre-flight
followed its own printed advice, reported OK, and the kernel build then died with
nvidia/cu13/include/crt/host_config.h:208: fatal error: features.h: No such file
16 times in one run. `features.h` is glibc's, not CUDA's. On the minimal Azure Linux core
image `gcc-c++` does not pull `glibc-devel`, whereas Fedora/RHEL's gcc-c++ does and Debian's
build-essential brings libc6-dev — which is why only this distro shows it.
Verified in the container before changing anything: `tdnf install -y glibc-devel` installs
glibc-devel-2.38-20.azl3 and produces /usr/include/features.h.
Added only where it is needed. Fedora and RHEL get it transitively, and a test asserts they
are not changed — every extra package in a command the user is asked to paste is noise, and
noise is how advice stops being read.
Not yet proven end to end: that an azurelinux install leg now completes. The package and the
header are confirmed; the full leg is for the next matrix run.
cnygaard
enabled auto-merge (squash)
August 19, 2026 19:27
This file contains hidden or bidirectional Unicode text that may be interpreted or compiled differently than what appears below. To review, open the file in an editor that reveals hidden Unicode characters.
Learn more about bidirectional Unicode characters
Sign up for free
to join this conversation on GitHub.
Already have an account?
Sign in to comment
Add this suggestion to a batch that can be applied as a single commit.This suggestion is invalid because no changes were made to the code.Suggestions cannot be applied while the pull request is closed.Suggestions cannot be applied while viewing a subset of changes.Only one suggestion per line can be applied in a batch.Add this suggestion to a batch that can be applied as a single commit.Applying suggestions on deleted lines is not supported.You must change the existing code in this line in order to create a valid suggestion.Outdated suggestions cannot be applied.This suggestion has been applied or marked resolved.Suggestions cannot be applied from pending reviews.Suggestions cannot be applied on multi-line comments.Suggestions cannot be applied while the pull request is queued to merge.Suggestion cannot be applied right now. Please check back later.
Found by the distro matrix, and it is precisely the failure mode this suite exists for.
Pre-flight followed its own printed advice, reported OK, and the kernel build then died with:
16 times in one run. Advice that leads to a state the user believes is ready is worse than no advice.
Why only Azure Linux
features.his glibc's, not CUDA's. On the minimal Azure Linux core image,gcc-c++does not pullglibc-devel. Fedora and RHEL get it transitively via theirgcc-c++; Debian gets it viabuild-essential→libc6-dev. So this is a one-distro gap, and it is fixed in one place.Verified before changing anything
Scope
Added only where needed. A test asserts fedora and rhel are not changed — every extra package in a command the user is asked to paste is noise, and noise is how advice stops being read.
Also worth recording: this is not a CUDA Toolkit problem
The same matrix run shows
Could not find nvcc0 times — nvcc ships with the pip CUDA wheels and was present on every distro and every Python version (3.11–3.14). The toolchain gaps found so far are distro dev packages, not NVIDIA's installer:unsupported GNU version(×105)features.h(×16)Not proven yet
That an azurelinux install leg now completes end to end. The package and the header are confirmed; the full leg is for the next matrix run.